[ python-Bugs-1467080 ] Many functions in socket module are not thread safe

[SourceForge.net]

Bugs item #1467080, was opened at 2006-04-08 18:09
Message generated for change (Comment added) made by sobomax
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=105470&aid=1467080&group_id=5470

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: None
Group: Python 2.4
Status: Open
Resolution: None
Priority: 5
Submitted By: Maxim Sobolev (sobomax)
Assigned to: Nobody/Anonymous (nobody)
Summary: Many functions in socket module are not thread safe

Initial Comment:
The socket module make a great effort to be thread-safe, but misses one 
big point - it uses single per-instance buffer to hold resolved 
sockaddr_xxx structures. Therefore, on SMP system it is possible that 
several threads calling functions that perform address resolution in 
parallel will stomp on each other resulting in incorrect or invalid address 
to be used in each case.

For example, two threads calling sendto() in parallel can result in packets 
to be sent to incorrect addresses - packets from thread one from time to 
time will be sent to address requested by thread two and vice versa.

Another, smaller issue is that the call to getnameinfo() is not protected 
with netdb mutex on systems that don't have thread-safe resolver.

----------------------------------------------------------------------

>Comment By: Maxim Sobolev (sobomax)
Date: 2006-04-17 13:26

Message:
Logged In: YES 
user_id=24670

> Why is it necessary to allocate the memory dynamically?
> Couldn't the caller provide the memory on the stack (i.e.
> through a local variable)?

Local variable is not very good IMHO since in threading 
environment with hundreds of threads running at the same 
time stack can be a scarce resource. Another issue is that 
the caller doesn't know the actual size it has to allocate 
until the resolution has been done, therefore it would need 
to overallocate in most cases.

-Maxim

----------------------------------------------------------------------

Comment By: Martin v. Löwis (loewis)
Date: 2006-04-10 22:22

Message:
Logged In: YES 
user_id=21627

Why is it necessary to allocate the memory dynamically?
Couldn't the caller provide the memory on the stack (i.e.
through a local variable)?

----------------------------------------------------------------------

Comment By: Maxim Sobolev (sobomax)
Date: 2006-04-10 17:08

Message:
Logged In: YES 
user_id=24670

Sorry, for some reason the patch did not go through first time. See attached.

----------------------------------------------------------------------

Comment By: Martin v. Löwis (loewis)
Date: 2006-04-09 14:18

Message:
Logged In: YES 
user_id=21627

I wonder why the sock_addr member is there in the socket
object in the first place. AFAICT, there is no need for it;
it was introduced in r4509.

Would you like to work on a patch?

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=105470&aid=1467080&group_id=5470